JP2013053875A - Inspection support device, program, and inspection support method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To reduce frequency of inspection and time and effort thereof by realizing a simple method for inspecting an abnormality (including deviation from a normality) by way of an appearance to support the inspection.SOLUTION: According to a vehicle inspection support system 1, a vehicle inspection support device 200 determines presence of an abnormality of a vehicle picked up in a photographed image on the basis of the photographed image taken by a photography device 100 disposed on a rail side. Specifically, the vehicle inspection support device 200 generates a vehicle image 30 of the entire one car by synthesizing individual frame images of a video-captured image and determines the presence of abnormality in the vehicle taken in the vehicle image 30 on the basis of the smallest relative distance of relative distances between the vehicle image 30 and determining-images that are past vehicle images in which the captures vehicles have been determined to be "normal."

Description

  The present invention relates to an inspection support apparatus and the like.

  For the purpose of regular inspection and maintenance, inspections such as work inspection and police box inspection are performed for railway vehicles, and vehicle inspection is performed for automobiles, but failures may occur before these inspections are reached. For example, in the case of a railway vehicle, the vehicle may cause an accident due to a failure, and there may be a suspension of service or a timetable disruption. However, since it is impossible to eliminate the failure, on the premise of the failure, measures such as ensuring fail-safe operation that operates safely even when a failure occurs and multiplexing parts are taken. .

  Various technologies for early detection of abnormalities leading to failures have been developed and put into practical use. For example, Patent Document 1 discloses a technique for accurately determining an abnormality in a bearing portion.

JP 2006-341659 A

  However, many of the conventional abnormality detection techniques are specialized techniques for detecting an abnormality of a specific part or a specific member. For this reason, it is not realistic to incorporate an individual detection system corresponding to a part into every part mounted on a railway vehicle or automobile.

  Even when the component itself is functioning normally, the outer shape is deformed or discolored, the fixture (fastening bolt, etc.) of the component is loose or damaged, and is not normally attached. In some cases, deposits (oil etc.) are attached. In many cases, such an abnormality, or a state that is not normal even if not abnormal, can be determined by appearance. Moreover, the above-mentioned problems apply to various mechanical products and electrical products in addition to railway vehicles and automobiles.

  The present invention has been made in view of the above-described problems, and an object of the present invention is to realize a technique for easily inspecting abnormalities (including being in a state different from normal) from the appearance, and to support the inspection. In other words, it contributes to increasing the frequency of inspections and omission of inspection labor.

The first form for solving the above problem is
Storage means (for example, the storage unit 230 in FIG. 8) storing a plurality of normal images of the inspection object determined to be normal;
Each of the captured images (for example, the vehicle image 30 in FIG. 5) obtained by capturing the inspection object to be inspected this time and one normal image among the plurality of normal images stored in the storage unit Inspection window initial setting means (for example, correlation distance calculation unit 222 in FIG. 8) for initial setting a predetermined size inspection window (for example, inspection window 40 in FIG. 5) at a predetermined relative position with respect to the object;
Gradually shifting the inspection window in a predetermined relative direction with respect to the inspection object, and comparing the images in the inspection window of each of the captured image and the one normal image to determine dissimilarity Image analysis means (for example, correlation distance calculation unit 222 in FIG. 8) that repeatedly executes
Non-similarity individual determination means (for example, the correlation distance in FIG. 8) in which the maximum dissimilarity among the non-similarities determined by the image analysis means is the dissimilarity of the captured image with respect to the one normal image. A calculation unit 222);
Of the dissimilarities determined by the dissimilarity individual determining means by switching the one normal image, causing the inspection window initial setting means, the image analyzing means and the dissimilarity individual determining means to repeatedly function. A final dissimilarity determination means (for example, the correlation distance calculation unit 222 in FIG. 8) that sets the minimum dissimilarity as the final dissimilarity of the captured image;
Based on the final dissimilarity determined by the final dissimilarity determining means, an abnormality determining means (for example, the abnormality in FIG. 8) for determining whether there is an abnormality in the current inspection object copied in the captured image. Determination unit 223),
Is an inspection support apparatus (for example, the vehicle inspection support apparatus 200 of FIG. 8).

As another form, a computer comprising storage means for storing a plurality of normal images of an inspection object determined to be normal,
Predetermined in a predetermined relative position with respect to the inspection object on each of the photographed image obtained by photographing the inspection object to be inspected this time and one normal image among the plurality of normal images stored in the storage means Inspection window initial setting means (for example, correlation distance calculation unit 222 in FIG. 8) for initial setting the size inspection window,
Gradually shifting the inspection window in a predetermined relative direction with respect to the inspection object, and comparing the images in the inspection window of each of the captured image and the one normal image to determine dissimilarity And an image analysis unit (for example, correlation distance calculation unit 222 in FIG. 8).
Non-similarity individual determination means (for example, the correlation distance in FIG. 8) in which the maximum dissimilarity among the non-similarities determined by the image analysis means is the dissimilarity of the captured image with respect to the one normal image. Calculation unit 222),
Of the dissimilarities determined by the dissimilarity individual determining means by switching the one normal image, causing the inspection window initial setting means, the image analyzing means and the dissimilarity individual determining means to repeatedly function. Final dissimilarity determination means (for example, the correlation distance calculation unit 222 in FIG. 8) that sets the minimum dissimilarity as the final dissimilarity of the captured image;
Based on the final dissimilarity determined by the final dissimilarity determining means, an abnormality determining means (for example, the abnormality in FIG. 8) for determining whether there is an abnormality in the current inspection object copied in the captured image. Determination unit 223),
It is good also as comprising the program (For example, the vehicle inspection assistance program 231 of FIG. 8) for functioning as.

Further, as another embodiment, an inspection support method for supporting an inspection of an inspection object by controlling a computer including a storage unit that stores a plurality of normal images of the inspection object determined to be normal,
Predetermined in a predetermined relative position with respect to the inspection object on each of the photographed image obtained by photographing the inspection object to be inspected this time and one normal image among the plurality of normal images stored in the storage means An inspection window initial setting step (for example, step A5 in FIG. 10) for initial setting the size inspection window;
Gradually shifting the inspection window in a predetermined relative direction with respect to the inspection object, and comparing the images in the inspection window of each of the captured image and the one normal image to determine dissimilarity An image analysis step (for example, step A7 in FIG. 10) for repeatedly executing
Dissimilarity individual determination step (for example, step S9 in FIG. 10) in which the maximum dissimilarity among the dissimilarities determined in the image analysis step is the dissimilarity of the captured image with respect to the one normal image. )When,
The one normal image is switched, the inspection window initial setting step, the image analysis step, and the dissimilarity individual determination step are repeatedly executed, and the dissimilarity determined in the dissimilarity individual determination step A final dissimilarity determination step (for example, the process of loop A in FIG. 10) that sets the minimum dissimilarity as the final dissimilarity of the captured image;
Based on the final dissimilarity determined in the final dissimilarity determining step, an abnormality determining step (for example, the step of FIG. 10) for determining whether or not there is an abnormality in the current inspection object copied in the captured image. S11 to S13),
It is good also as comprising the test | inspection assistance method containing these.

  According to the first embodiment and the like, each of the photographed image of the inspection object to be inspected this time and one normal image of the inspection object has a predetermined size at a predetermined relative position with respect to the inspection object. The inspection window is initialized. Then, the inspection window is gradually shifted in a predetermined relative direction, and the images in the inspection window are compared with each other to determine the dissimilarity. Among the determination results, the maximum dissimilarity is the dissimilarity of the captured image with respect to one normal image.

  That is, since the inspection window is initially set at a predetermined relative position with respect to the inspection object and is shifted in a predetermined relative direction, the determination of the dissimilarity between the images in the inspection window is more accurate. Will be done. Of the determined dissimilarities, the maximum dissimilarity is the dissimilarity of the captured image with respect to one normal image. As a result, of the portions in the inspection window, the portion that is most similar to the normal image is clarified. This is an individual determination for one normal image of the captured image.

  Furthermore, according to the first embodiment, the individual determination from the initial setting of the inspection window to the shift of the inspection window and the determination of the dissimilarity is repeated by switching one normal image and individually determined. The minimum dissimilarity among the degrees is set as the final dissimilarity of the captured image. Since there are a plurality of normal images, the normal image that is the closest, that is, the smallest dissimilarity is the closest to the captured image of the current inspection object. However, although it is the most approximate, the dissimilarity of the least similar part in the normal image is the final dissimilarity. As a result, based on the final dissimilarity, it is possible to determine whether there is an abnormality in the current inspection object.

  According to the above configuration, as long as a captured image of the current inspection object can be acquired, it is possible to easily determine whether there is an abnormality in the current inspection object. As a result, it is possible to contribute to omission of inspection and high-frequency inspection.

As a second form, in the first form,
An abnormality determination position for controlling the display of the captured image and identifying and displaying the position of the inspection window in the final captured image that is set to the final dissimilarity when the abnormality determination unit determines that there is an abnormality. It is good also as comprising the examination support device further provided with the display control means (for example, information control part 225 of Drawing 8).

  According to the second embodiment, the display of the photographed image is controlled, and the position of the examination window determined as the final dissimilarity in the photographed image is identified and displayed. Helps to make judgments.

As a third form, in the first or second form,
An imaging control means (for example, FIG. 1) that is installed toward the passing point of the inspection object and that continuously captures a part of the inspection object as an imaging range according to the passage of the inspection object, thereby capturing the entire passing direction of the inspection object. The whole image generating means (for example, the vehicle image creating unit 221 in FIG. 8) that generates, as the photographed image, one whole image in which the entire passing direction of the inspection object is settled from the images continuously photographed by the photographing device 100). It is good also as comprising the inspection assistance apparatus further provided with.

  According to the third aspect, the imaging control means installed toward the passing point of the inspection target object continuously inspects the inspection object as a part of the inspection object according to the passage of the inspection object. The entire direction of passage is taken. Then, from the continuously photographed images, one whole image in which the entire passing direction of the inspection object is contained is generated as a photographed image for determining the dissimilarity. Therefore, it is not necessary to set and set the shooting range of the shooting control means to an angle of view that allows the entire inspection object to be shot, so that the shot image can be made as high-definition as possible. In addition, since one whole image in which the entire passing direction of the inspection object is contained is used as a photographed image for determining dissimilarity, initial setting of the inspection window and determination of the shift direction of the inspection window are facilitated. .

As a fourth form, in any one of the first to third forms,
The storage means stores normal images of a plurality of types of inspection objects including the same type as the current inspection object,
The final dissimilarity is determined by determining the maximum dissimilarity by the individual dissimilarity determining unit and determining the minimum dissimilarity as the final dissimilarity by the final dissimilarity determining unit. It is also possible to configure an inspection support apparatus characterized in that the type of normal image according to the degree automatically becomes the same type as the current inspection object.

  According to the fourth embodiment, since there is no trouble of inputting the type of the inspection object this time, it is possible to more easily realize the abnormality determination.

The block diagram of a vehicle inspection assistance system. Explanatory drawing of producing | generating a vehicle image by synthesize | combining a picked-up image. An example of a set of vehicle images including a plurality of types of trains. The result of applying hierarchical clustering to the set of vehicle images in FIG. Explanatory drawing of the setting of the inspection window with respect to a vehicle image. Relationship between the position of the inspection window and the correlation coefficient between the partial images in the inspection window. The result of applying hierarchical clustering using the distance between images calculated using the inspection window. The function block diagram of a vehicle inspection assistance apparatus. The data structural example of image DB for determination. The flowchart of a vehicle inspection assistance process.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings. However, the applicable embodiment of the present invention is not limited to this.

[system]
FIG. 1 is an overall configuration diagram of a vehicle inspection support system 1 according to the present embodiment. As shown in FIG. 1, the vehicle inspection support system 1 includes an imaging device 100 and a vehicle inspection support device 200.

  The imaging device 100 is installed beside the track and images the train 10 traveling on the track from the side. The photographing apparatus 100 is a digital camera or a digital video camera capable of photographing a moving image by high-speed continuous shooting, and includes the entire height direction of the train 10 as an inspection target in the photographing range and the traveling direction of the train 10. Is set and arranged so that the angle of view and the shooting direction are set so as to match the horizontal direction of the shot image. Note that the entire longitudinal direction of the train 10 need not be included in the imaging range. Moreover, the imaging device 100 has a built-in communication device, and the captured image data can be transmitted to the vehicle inspection support device 200 by this communication device.

  The vehicle inspection support device 200 is realized by a computer system, and is installed in, for example, a management center. The vehicle inspection support device 200 performs processing for supporting the abnormality determination of the train 10 based on the captured image received from the imaging device 100.

[principle]
The principle of the vehicle inspection support process in this embodiment will be described.

(1) Image Synthesis First, a vehicle image 30 in which an entire vehicle is shown is created from a photographed image 20 taken by the photographing apparatus 100. Railway vehicles are elongated in the lateral direction (traveling direction). Since the photographing device 100 installed on the side of the track shoots the train 10 from a close range, it is difficult to fit the entire vehicle into one image. Rather, in order to finally obtain a highly accurate photographed image, An angle of view and an arrangement position are determined so as to photograph a part. By combining the still images taken continuously (moving image shooting) by the photographing apparatus 100, a vehicle image 30 that is an entire image of one vehicle is created.

  FIG. 2 is a diagram for explaining creation of a vehicle image by image composition. As shown in FIG. 2, each captured image 20 shows a part of the vehicle. One vehicle image 30 is created by combining the plurality of captured images 20. That is, a common part of two photographed images 20 adjacent to each other in the photographing order is determined and combined so that the common parts overlap each other, so that one image is finally obtained.

  Specifically, the position of the imaging device 100 is fixed, and the traveling direction of the train 10 coincides with the horizontal direction of the captured image 20, so that the two captured images 20 are overlapped in the vertical direction, and the horizontal direction What is necessary is just to look for the position where the overlapping image parts correspond, shifting. Further, the matching position of the “image part” can be determined from the fact that the correlation coefficient between the image parts is maximized. The correlation coefficient between images is a Pearson product-moment correlation coefficient, and is calculated as a correlation coefficient using a set of color data (numerical data representing colors) of each pixel constituting each image as a variable. However, the correlation coefficient in this embodiment is 0 or more and 1 or less.

(2) Basic concept of abnormality determination Once the vehicle image 30 is created (synthesized), the vehicle image 30 is used to determine whether there is an abnormality in the vehicle that is shown. Specifically, the current vehicle image 30 is compared with the past “normal” vehicle image 30, and when both are significantly different, it is determined that the train shown in the current vehicle image 30 has “abnormal”. To do.

In addition, as a reference for comparison between the vehicle images 30, “dissimilarity” indicating how different the images are is used. This dissimilarity between images is represented by “correlation distance d” defined by the following equation (1).
Correlation distance d = 1-correlation coefficient c (1)

  The presence / absence of abnormality of the train shown in the image is determined using the correlation coefficient between the vehicle images 30, and there are a plurality of types (types) of vehicles constituting the train. That is, even if there is no abnormality in the train, if the vehicle type is different, each vehicle image 30 is greatly different (that is, the “correlation distance d” between the images is long). For this reason, it is necessary to discriminate the vehicle type shown in the vehicle image 30 and to compare between the vehicle images 30 showing the trains of the same vehicle type.

  Therefore, in the present embodiment, vehicle type discrimination and abnormality determination are realized using the concept of “hierarchical clustering”. The hierarchical clustering used here is a pseudo-collective method, and the distance calculation is a single connection method (shortest distance method). The basic concept of vehicle type discrimination and abnormality determination using this hierarchical clustering will be described.

  First, a “cluster” including each of the plurality of vehicle images 30 is generated. That is, the same number of clusters as the vehicle image 30 are generated. Next, the “distance” between the clusters is calculated, and the two clusters with the shortest distance are combined into one cluster. Here, the distance between the clusters is the distance between the captured images 30 included in the clusters. The distance between the captured images 30 is the correlation distance d described above.

  The process of combining two clusters having the shortest distance among the distances between the clusters is repeated until one cluster is finally obtained. At this time, regarding the distance between the clusters including two or more captured images 30, the shortest distance among the distances between the captured images 30 included in each cluster is defined as the distance between the clusters. As a result, a hierarchical structure of a binary tree that hierarchically represents how the captured images 30 are combined is obtained.

  FIG. 3 is a diagram illustrating an example of a set of vehicle images 30, and FIG. 4 is a diagram illustrating an example of a result of applying hierarchical clustering to the set of vehicle images 30 in FIG. FIG. 3 shows a set of 14 vehicle images 30 (images 1 to 14). Two types of vehicles are shown in these images 1 to 14, images 1 to 6 and 14 are “type A”, and images 7 to 13 are “type B”. Images 1 to 13 show “normal” vehicles, and image 14 shows “abnormal” vehicles.

  FIG. 4 is a dendrogram (dendrogram) showing a hierarchical structure obtained by hierarchical clustering. In this tree diagram, “leaf” represents each of the vehicle images 30 (images 1 to 14), and the height from this “leaf” to the combined position is that when the vehicle image 30 is combined with another cluster. This represents the distance (correlation distance d). Moreover, the distance between clusters is represented by the height to the joint position of the clusters. For example, the distance between the image 1 and the image 2 (correlation distance d) is the height H1, and the distance between the cluster including the images 1 and 2 and the cluster including the image 14 is the height H2.

  The set of vehicle images shown in FIG. 3 includes images showing two types of vehicles. As shown in the tree diagram of FIG. 4, the two clusters are finally combined into one cluster, but these two clusters may be separated by the vehicle type of the captured image 30 included. Recognize. That is, a cluster including images 1 to 6 and 14 showing the “type A” vehicles (left side in the figure) and a cluster including images 7 to 13 showing the “type B” vehicles (right side in the figure). is there. In addition, the distance (height) between the two clusters is extremely longer than the distance between the lower level clusters, and it can be automatically determined that the vehicle type is different between the two clusters. .

  Further, the length of each image branch (height to the first combining position: for example, H1 in the case of image 1) is the dissimilarity between the image and the closest other image among all the images. It can be considered that the length of the branch is long, that is, the degree of dissimilarity is large, that is, it is greatly different (abnormal) from other images. For example, in FIG. 4, the length of the branch of the image 14 is the longest among all the images 1 to 14. That is, it can be determined that the vehicle shown in the image 14 is significantly different (abnormal) from other vehicles.

(3) Abnormality determination method of the present embodiment As described above, by calculating the correlation distance d between the vehicle images 30, it is possible to determine the abnormality of the reflected train. However, for example, as shown in an image 14 in FIG. 3, the ratio of “abnormal spots” of the train to the entire image is small. For this reason, the similarity of the vehicle image 30 as a whole is relatively high. For example, as shown in the tree diagram shown in FIG. 4, the branch of the image 14 having the “abnormality” is the longest, but the overall branches of the other “normal” images 1 to 13 are viewed as a whole. The difference with the length of is not very large.

  Therefore, in the present embodiment, as shown in FIG. 5, by setting an “inspection window (inspection region)” in the vehicle image, a comparison focusing on an abnormal portion in the image rather than the entire vehicle image is performed.

  FIG. 5 is a diagram illustrating the setting of the “inspection window” in the vehicle image. FIG. 5 shows two vehicle images 30-1 and 30-2 in which the same type of vehicle is copied. One vehicle image 30-1 is an image of a “normal” vehicle, and the other vehicle image 30-2 is an image of a vehicle having an “abnormal”. Specifically, there is an abnormal location 12 in the underfloor device near the vehicle center in the vehicle image 30-2. As shown in FIG. 5, an inspection window 40 having the same size is set for each of the two vehicle images 30. Further, the inspection window 40 is set at the same relative position with respect to the vehicle in the image. For example, the position in the vehicle image 30 of a predetermined part of the vehicle (for example, the leading position of the vehicle or the end of the frontmost door in the traveling direction in the vehicle) is specified, and this position is used as a reference. The inspection window 40 is set.

  Then, while gradually shifting the inspection window in each vehicle image 30 at the same speed in the same direction (for example, in the right direction), a position where the correlation coefficient between the partial images in the inspection window 40 is minimized is searched. The correlation distance between the partial images in the inspection window 40 at the position is a correlation distance d between the vehicle images 30.

  Here, the inspection window 40 is set and moved so as to include an abnormality determination target portion in the vehicle image 30. In the present embodiment, an underfloor device of a vehicle is a target for abnormality determination. For this reason, let the length of the height direction of the inspection window 40 be the length that the underfloor equipment can be accommodated. Specifically, the inspection window 40 is initially set to an initial position P2 that is determined with reference to the leading position P1 of the vehicle shown in each vehicle image 30. The initial position P1 is a position in front of the underfloor device to be determined. Then, the inspection window 40 is moved in the lateral direction (the direction opposite to the traveling direction). Note that the position in the vertical direction (height direction) of the vehicle image 30 that defines the inspection window 40 is constant if the installation position of the imaging device 100 is fixed. However, the vertical position may of course be initialized.

  FIG. 6 is a diagram illustrating a correlation coefficient of the partial image in the inspection window 40 with respect to the vehicle image 30. In FIG. 6, two vehicle images 30-1 and 30-2 are shown on the upper side, and the correlation coefficient of the partial images in the inspection window 40 is shown on the lower side. According to FIG. 6, it can be seen that the correlation coefficient is rapidly decreased at a position including the abnormal portion 12 of the vehicle in the inspection window 40. The correlation coefficient is relatively large at positions other than the vicinity of the abnormal portion 12.

  FIG. 7 shows the result of applying hierarchical clustering to the set of vehicle images 30 shown in FIG. 3, and shows a case where the correlation distance between the vehicle images 30 is calculated using the inspection window 40. In the tree diagram shown in FIG. 7, the length of each image branch (correlation distance d) is longer than that in the tree diagram shown in FIG. 4. That is, the difference between the vehicle images 30 is emphasized, and it can be said that the accuracy of the abnormality determination of the vehicle in the image is improved.

  In the tree diagram in FIG. 7, as in the tree diagram shown in FIG. 4, finally, two clusters divided by vehicle type (images 1 to 6 and 14 showing vehicles of “type A” are shown. (The left side in the figure) and the cluster including the images 7 to 13 depicting the “type B” vehicle (the right side in the figure) are combined into one cluster. However, the hierarchical structure included in each of the two clusters divided by vehicle type is different from the tree diagram shown in FIG. This is because, as a result of comparison in a small range of image portions in the inspection window 40, not in the entire vehicle, a difference occurs in the correlation distance, and the cluster combination order is changed. It can be said that the accuracy of abnormality determination has improved.

  Further, in the tree diagram shown in FIG. 7, the length of the branch of the image 14 (correlation distance d) is the longest as in the tree diagram shown in FIG.

  As described above, in this embodiment, it is possible to determine the abnormality of the train shown in the vehicle image by using the hierarchical clustering and using the correlation distance d between the vehicle images.

[Function configuration]
FIG. 8 is a block diagram illustrating a functional configuration of the vehicle inspection support device 200. As shown in FIG. 8, the vehicle inspection support apparatus 200 functionally includes an operation unit 211, a communication unit 212, a display unit 213, a sound output unit 214, a processing unit 220, and a storage unit 230. It is prepared for.

  The operation unit 211 is realized by an input device such as a button switch, lever, keyboard, mouse, touch panel, or various sensors, and outputs an operation signal corresponding to the operation instruction to the processing unit 220.

  The communication unit 212 is realized by a communication device such as a wireless communication module, a router, a modem, a TA, or a wired cable jack, and realizes communication with an external device (mainly, the imaging device 100).

  The display unit 213 is realized by a display device such as a CRT, LCD, ELD, or PDP, and displays a display screen based on a display signal from the processing unit 220.

  The sound output unit 214 is realized by a sound output device such as a speaker, and outputs sound based on the sound signal from the processing unit 220.

  The processing unit 220 is realized by an arithmetic device such as a CPU, for example, and is based on a program or data stored in the storage unit 230, an operation signal from the operation unit 211, received data from the communication unit 212, or the like. An instruction and data transfer to each part constituting 200 are performed, and overall control of the vehicle inspection support device 200 is performed. The processing unit 220 includes a vehicle image creation unit 221, a correlation distance calculation unit 222, an abnormality determination unit 223, a determination image management unit 224, and a notification control unit 225 according to the vehicle inspection support program 231. Then, a support process for supporting vehicle abnormality determination is performed based on a photographed image by the photographing apparatus 100.

  The vehicle image creation unit 221 synthesizes the captured image 20 by the photographing apparatus 100 and creates a vehicle image 30 in which one entire vehicle is shown. Here, data relating to the captured image 20 is stored as captured image data 235, and data relating to the created vehicle image 30 is stored as vehicle image data 236.

  The correlation distance calculation unit 222 calculates a correlation distance d between the vehicle image 30 created by the vehicle image creation unit 221 and each of the determination images. Specifically, the inspection window 40 of a predetermined size is set so that the relative position with respect to the vehicle in the vehicle image 30 and the determination image coincide with each other. Then, by moving the set inspection window 40 at the same speed in the same direction, the position of the inspection window 40 at which the correlation coefficient between the partial images in the inspection window 40 is minimized is searched, and the inspection window 40 at that position is searched. The correlation distance between the partial images is a correlation distance d between the vehicle image 30 and the determination image. Here, the determination image is a past vehicle image 30 that has been determined to be “normal”, and a large number of corresponding data is stored in the determination image DB 233.

  The abnormality determination unit 223 selects the minimum correlation distance from the correlation distances between the vehicle image 30 and all of the determination images calculated by the correlation distance calculation unit 222, and selects this as the vehicle image 30. Is the correlation distance d. This minimum correlation distance corresponds to the length of a branch in the tree diagram obtained by applying hierarchical clustering. Next, it is determined whether or not the vehicle shown in the vehicle image 30 is abnormal depending on whether or not the selected correlation distance d exceeds a predetermined threshold distance. The threshold distance serving as the determination criterion at this time is determined as, for example, the maximum of the correlation distances between the plurality of “normal” vehicle images (determination images). This is because it is determined as “abnormal” when the correlation distance between the “normal” vehicle images is exceeded. The maximum correlation distance among the correlation distances between the “normal” vehicle images is not set as the threshold distance, but a value obtained by adding a predetermined value to the maximum correlation distance is set as the threshold distance, The threshold distance may be 1.2 times the correlation distance. The threshold distance is stored in advance as abnormality determination threshold data 232.

  The notification control unit 225 controls notification of the determination result by the abnormality determination unit 223. Specifically, for example, a predetermined notification screen on which a determination result (normal / abnormal) is displayed together with the vehicle image 30 to be determined is displayed on the display unit 213, or a predetermined notification sound corresponding to the determination result is displayed. The sound is output from the sound output unit 214. Furthermore, when the determination result is “abnormal”, the position of the inspection window 40 when the correlation coefficient c with the determination image is minimized is displayed in an overlapping manner on the vehicle image 30. , It is possible to show an abnormal part of the vehicle being reflected.

  The determination image management unit 224 manages the determination image DB 233. Specifically, when the determination result by the abnormality determination unit 223 is “normal”, the vehicle image 30 to be determined is added to the determination image DB 233 as a new determination image. As a result, if the number of managed judgment images exceeds the predetermined number, the judgment image having the smallest correlation distance is deleted from the judgment image DB 233 among all the judgment images. Specifically, since hierarchical clustering is applied, there are two determination images with the smallest correlation distance, and these two determination images are the most similar among all the determination images. It is an image. Therefore, it can be considered that even if one of the two determination images is deleted, the abnormality determination is not affected. Further, of the two determination images, for example, the one with the oldest shooting date and time may be deleted.

  FIG. 9 is a diagram illustrating an example of a data configuration of the determination image DB 233. As shown in FIG. 9, the determination image DB 233 is a set (database) of determination image data 234 for each managed determination image, and each determination image data 234 is an image No. that is an identification number. In association with 234a, a determination image 234b, a shooting date and time 234c, and a correlation distance 234d are stored. The shooting date / time 234c is the shooting date / time of the shot image that is the basis of the corresponding determination image (vehicle image). The correlation distance 234d corresponds to the branch length of the corresponding determination image in the tree diagram obtained by applying hierarchical clustering.

  The storage unit 230 is realized by a storage device such as a hard disk, a ROM, or a RAM, and stores a system program for the processing unit 220 to control the vehicle inspection support device 200 in an integrated manner, a program and data for realizing various functions. In addition to being stored, it is used as a work area for the processing unit 220, and the results of operations performed by the processing unit 220 for various processes are temporarily stored in an image. In the present embodiment, the storage unit 230 stores a vehicle inspection support program 231 as a program, and includes as data abnormality determination threshold data 232, determination image DB 233, captured image data 235, and vehicle image data 236. Is memorized.

[Process flow]
FIG. 10 is a flowchart for explaining the flow of the vehicle inspection support process. According to FIG. 10, when the captured image 20 by the imaging device 100 is acquired (step S1), the vehicle image creation unit 221 combines the captured image 20 to create a vehicle image 30 (step S3). Next, loop A processing (repetition processing) is performed on each managed determination image.

  In the loop A, the correlation distance calculation unit 222 sets the inspection window 40 of the same size so that the relative position with respect to the vehicle in the vehicle image 30 and the target determination image match each other (step S5). ). Next, the inspection window 40 in both images is moved in the same direction at the same speed, and a position where the correlation coefficient between the partial images in the inspection window 40 is minimized is searched (step S7). Then, the correlation distance between the partial images in the inspection window 40 at this position is calculated and set as the correlation distance d (dissimilarity) between the vehicle image 30 and the target determination image (step S9). That is, the maximum value of the correlation distances between the partial images in the inspection window 40 is set. The process of loop A is performed in this way.

  Then, the processing of loop A for all the determination images is performed to calculate the correlation distance d between the vehicle image 30 and each of the determination images, and then the abnormality determination unit 223 performs these determinations. The minimum distance (minimum dissimilarity) is selected from the correlation distances d between the images for use, and is set as the correlation distance of the vehicle image 30 (step S11).

  Next, the correlation distance of the vehicle image 30 is compared with a predetermined threshold distance. If the correlation distance exceeds the threshold distance (step S13: YES), the vehicle shown in the vehicle image has “abnormal”. Determination is made (step S15). And the alerting | reporting control part 225 alert | reports a determination result (step S17).

  On the other hand, if the correlation distance does not exceed the threshold distance (step S13: NO), the abnormality determination unit 223 determines that the vehicle shown in the vehicle image is “normal” (step S19). And the alerting | reporting control part 225 alert | reports a determination result (step S21).

  Subsequently, the determination image management unit 224 adds the vehicle image 30 to the determination image DB 233 as a new determination image (step S23). As a result, if the number of managed determination images exceeds the predetermined upper limit number (step S25: YES), among the managed determination images, the correlated distance that is associated is the smallest. Are deleted from the determination image DB 233 (step S27). When the above processing is performed, the processing unit 220 ends the support processing.

[Action / Effect]
As described above, according to the vehicle inspection support system 1 of the present embodiment, a train as an inspection object is imaged (moving image shooting) by the imaging device 100 installed beside the track, and the vehicle inspection support device 200 performs this imaging. It is possible to determine the presence or absence of an abnormality that can be determined from the appearance of the vehicle shown in the photographed image by the apparatus 100.

  Specifically, the vehicle inspection support device 200 generates a vehicle image 30 of one entire vehicle by combining the frame images of the images captured by the imaging device 100 (moving image capturing). Whether the vehicle shown in the vehicle image 30 is abnormal based on the correlation distance d between each of the vehicle images shown in FIG. Determine whether or not. In addition, there are a plurality of types of trains traveling on the track, but whether or not the abnormality is based on the minimum correlation distance among the correlation distances representing the dissimilarity between the vehicle image and each of the determination images. Therefore, as a result, since it is compared with a determination image in which a vehicle of the same vehicle type as the vehicle image is captured, it is possible to easily determine the abnormality of the vehicle reflected in the vehicle image.

  When calculating the correlation distance d between the vehicle image 30 and the determination image, the inspection window 40 of a predetermined size is located at the same relative position with respect to the vehicle in the vehicle image 30 and the determination image. And the correlation distance between the partial images in the inspection window 40 at the position where the correlation coefficient between the partial images in the inspection window 40 is minimum (that is, the position where the correlation distance representing the dissimilarity is maximum). Is calculated as the correlation distance between the vehicle image 30 and the determination image. As a result, the “abnormal part” of the vehicle image 30 is clarified and the accuracy of the abnormality determination based on the vehicle image 30 is improved.

  Furthermore, when it is determined that there is an abnormality, the inspector can easily display the abnormality of the vehicle by displaying the position of the inspection window 40 that minimizes the correlation coefficient on the vehicle image 30 as a notification of the determination result. The location can be grasped.

[Modification]
It should be noted that embodiments to which the present invention can be applied are not limited to the above-described embodiments, and can be appropriately changed without departing from the spirit of the present invention.

(A) Shooting range of inspection object For example, in the above-described embodiment, it is decided to shoot from the side so that the entire height direction of the vehicle is included in the shooting range. It only has to be able to shoot to include. That is, if the underfloor device is the target of abnormality determination as in the present embodiment, it is sufficient to photograph at least the lower half of the vehicle in the photographing range. For example, when an apparatus on the roof is a target for abnormality determination, the upper half of the vehicle may be imaged so as to be included in the imaging range, or the image may be taken from above the vehicle. Further, in the vehicle inspection site, a photographing device may be installed at a position where photographing can be performed from below the vehicle, and photographing may be performed so that the underfloor equipment is photographed from below the vehicle.

(B) Inspection object Moreover, in the above-mentioned embodiment, although the railway vehicle was mentioned as an example as an inspection object, it is applicable similarly, for example to a motor vehicle. In that case, for example, the inside of the engine room may be the target of abnormality determination, or the bottom surface of the vehicle may be the target of abnormality determination. Further, the present invention is not limited to a vehicle, and can be similarly applied to any object that can be moved or conveyed and that can be determined from an appearance, such as a product inspection line in a factory.

(C) Trimming / masking of image In addition, since the background other than the vehicle is also reflected in the vehicle image 30, only the image portion of the vehicle is extracted and the remaining image portion is deleted (no color data is included). In the above, abnormality determination may be performed.

DESCRIPTION OF SYMBOLS 1 Vehicle inspection support system 100 Imaging device 200 Vehicle inspection support device 211 Operation part, 212 Communication part, 213 Display part, 214 Sound output part 220 Processing part 221 Vehicle image creation part, 222 Correlation distance calculation part 223 Abnormality determination part, 224 determination Image management unit, 225 notification control unit 230 storage unit 231 vehicle inspection support program 232 abnormality determination threshold data 233 determination image DB
235 Photographed image data, 236 Vehicle image data 10 Train 20 Photographed image, 30 Vehicle image, 40 Inspection window

Claims (6)

Storage means for storing a plurality of normal images of inspection objects determined to be normal;
Predetermined in a predetermined relative position with respect to the inspection object on each of the photographed image obtained by photographing the inspection object to be inspected this time and one normal image among the plurality of normal images stored in the storage means Inspection window initial setting means for initial setting of the inspection window of the size;
Gradually shifting the inspection window in a predetermined relative direction with respect to the inspection object, and comparing the images in the inspection window of each of the captured image and the one normal image to determine dissimilarity Image analysis means for repeatedly executing
A non-similarity individual determination unit that sets a maximum non-similarity among the non-similarities determined by the image analysis unit as a non-similarity of the captured image with respect to the one normal image;
Of the dissimilarities determined by the dissimilarity individual determining means by switching the one normal image, causing the inspection window initial setting means, the image analyzing means and the dissimilarity individual determining means to repeatedly function. A final dissimilarity determination means that sets a minimum dissimilarity as a final dissimilarity of the captured image;
An abnormality determining means for determining whether or not there is an abnormality in the current inspection object copied in the captured image based on the final dissimilarity determined by the final dissimilarity determining means;
An inspection support apparatus comprising:   An abnormality determination position for controlling the display of the captured image and identifying and displaying the position of the inspection window in the final captured image that is set to the final dissimilarity when the abnormality determination unit determines that there is an abnormality. The inspection support apparatus according to claim 1, further comprising display control means.   It was installed toward the inspection object passage point, and was continuously photographed by the photographing control means for photographing the entire passage direction of the inspection object by continuously photographing a part of the inspection object as the photographing range according to the passage of the inspection object. The inspection support apparatus according to claim 1, further comprising an entire image generation unit configured to generate, as the captured image, one entire image in which the entire passing direction of the inspection object is contained from the image. The storage means stores normal images of a plurality of types of inspection objects including the same type as the current inspection object,
The final dissimilarity is determined by determining the maximum dissimilarity by the individual dissimilarity determining unit and determining the minimum dissimilarity as the final dissimilarity by the final dissimilarity determining unit. The inspection support apparatus according to any one of claims 1 to 3, wherein the type of the normal image according to the degree automatically becomes the same type as the current inspection object. A computer having storage means for storing a plurality of normal images of the inspection object determined to be normal,
Predetermined in a predetermined relative position with respect to the inspection object on each of the photographed image obtained by photographing the inspection object to be inspected this time and one normal image among the plurality of normal images stored in the storage means Inspection window initial setting means for initial setting of inspection window of size,
Gradually shifting the inspection window in a predetermined relative direction with respect to the inspection object, and comparing the images in the inspection window of each of the captured image and the one normal image to determine dissimilarity And image analysis means for repeatedly executing.
A non-similarity individual determination unit that sets a maximum non-similarity among the non-similarities determined by the image analysis unit as a non-similarity of the captured image with respect to the one normal image;
Of the dissimilarities determined by the dissimilarity individual determining means by switching the one normal image, causing the inspection window initial setting means, the image analyzing means and the dissimilarity individual determining means to repeatedly function. A final dissimilarity determination means that sets a minimum dissimilarity as a final dissimilarity of the captured image;
An abnormality determining means for determining whether or not there is an abnormality in the current inspection object copied in the captured image, based on the final dissimilarity determined by the final dissimilarity determining means;
Program to function as. An inspection support method for supporting an inspection of an inspection object by controlling a computer including a storage unit that stores a plurality of normal images of the inspection object determined to be normal,
Predetermined in a predetermined relative position with respect to the inspection object on each of the photographed image obtained by photographing the inspection object to be inspected this time and one normal image among the plurality of normal images stored in the storage means An inspection window initial setting step for initial setting of an inspection window of a size;
Gradually shifting the inspection window in a predetermined relative direction with respect to the inspection object, and comparing the images in the inspection window of each of the captured image and the one normal image to determine dissimilarity And an image analysis step for repeatedly executing.
A non-similarity individual determination step in which the maximum non-similarity among the non-similarities determined in the image analysis step is a non-similarity of the captured image with respect to the one normal image;
The one normal image is switched, the inspection window initial setting step, the image analysis step, and the dissimilarity individual determination step are repeatedly executed, and the dissimilarity determined in the dissimilarity individual determination step A final dissimilarity determination step in which a minimum dissimilarity is set as a final dissimilarity of the captured image;
Based on the final dissimilarity determined in the final dissimilarity determining step, an abnormality determining step for determining whether there is an abnormality in the current inspection object copied in the captured image;
Inspection support method including

Images